ME-PFP: An Ensemble Learning Approach Fusing Multi-Source Features for Protein Function Prediction.

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Proteins are crucial in biological systems and are key targets in drug discovery and disease research.
• Accurate protein function prediction is vital but challenged by data integration and heterogeneous feature utilization.
• Existing methods often underutilize protein data, limiting prediction accuracy.

Purpose of the Study:

• To develop an advanced computational framework for enhanced protein function prediction.
• To address limitations in current methods regarding data integration and heterogeneous feature fusion.
• To improve the accuracy and efficiency of predicting protein functions for biological and medical applications.

Main Methods:

• Proposed ME-PFP, a novel ensemble learning framework for protein function prediction.
• Integrated sequence representations from protein language models, domain information, and protein-protein interaction data.
• Employed specialized attention-based feature extractors and a dynamic weighting strategy for effective heterogeneous data fusion.

Main Results:

• ME-PFP demonstrated significant improvements over existing sequence-based and multisource fusion models.
• Achieved an average accuracy improvement of 13.23% on human datasets and 11.11% on yeast datasets.
• The framework effectively captured and utilized heterogeneous features, enhancing prediction performance.

Conclusions:

• The ME-PFP framework offers a superior approach to protein function prediction.
• This advancement improves accuracy in computational biology and aids drug discovery research.
• The study highlights the potential of integrating diverse data modalities for biological predictions.